Positive germline selection of mtDNA Mutations: evidence from the oocyte

Purifying selection of mtDNA mutations is a vital process that cleanses the mitochondrial genome of detrimental variants that may endanger individuals and populations. A common measure of purifying selection is the increase of average synonymity by reduc tion the proportion of mostly detrimental non -synonymous mutations. The mechanisms underlying purifying selection are still debated. The Makova group has recently published high - fidelity analysis of mtDNA mutations in individual human oocytes (Arbeithuber et al., 2025). The authors observed a decrease in the proportion of potentially detrimental coding and conservative mutations at high er mutant fractions (MFs) and interpreted this as purifying selection removing detrimental mutations at higher MFs. We noted, however, that, in contrast to what would be expected under purifying selection, the synonymity of oocyte mutations was very low and decreased, rather than increased, at higher MFs. We hypothesized that this inconsistency resulted from non -synonymous mutations being prone to strong positive selection which erroneously made coding mutations appear negatively selected in comparison. In support of our hypothesis, we show that non-coding oocytes mutations indeed are under strong positive selection. To alleviate this setback, we reanalyzed the data using a new metric of intracellular clonal selection and neutral synonymous mutations as the reference. We demonstrated that coding mutations are in fact under prevailing positive selection. This is in line w ith previous estimates of positive selection in primordial germ cells (PGCs) and in mother-child pairs. Importantly, “prevailing positive selection” does not imply the absence of negative selection. We show that specific types of mutations may be under prevailing purifying selection (e.g., the Co1 gene). Of note, this prevailing positive selection pertains only to the most recent, germline mtDNA mutations which have not been yet inherited into the next generation. Purifying selection steps in as germline mutations proceed to subsequent generations. The imp lications of these findings and the potential benefits of positive selection of detrimental mtDNA mutations are discussed.

The fundamental role and mechanisms of germline selection of mtDNA have been the focus of mtDNA research for decades. Nascent mtDNA mutations are mostly nonsynonymous (synonymity ~0.2, i.e. 1 synonymous mutation per 4 nonsynonymous), whereas ‘old’ homolplasmic mtDNA mutations in population (such as mutations in deep branches of the human phylogenic tree) are ~ 0.75 synonymous (3 synonymous per 1 nonsynonymous). Thus, a substantial selective removal of nonsynonymous mutations, ‘purifying selection’, must drive this dramatic enrichment. Synonymity of course is just a simple way of monitoring selection, it can be generalized to various conservation scores. An important question, as far as germline selection is concerned, is what portion of this selection pro cess occurs in the germline, before Darwinian selection enters at the level of an individual’s survival or reproductive success. Selection is not limited to purifying (negative), however. Obviously, there is also positive selection of beneficial mutations. Intriguingly, also, there are cases of ‘selfish’ or ‘destructive’ positive selection of detrimental mutations. In somatic cells/tissues, positive selection has been observed in cardiomyocytes (Nekhaeva et al., 2002), muscle (Bua et al., 2006), (Nicholas et al., 2009), s. nigra (Kraytsberg et al., 2006), cancer (Yuan et al., 2020), fibroblasts (Korotkevich et al., 2025), to menti on just a few studies. Most puzzling, positive selection of detrimental mutations apparently may act in the germline during transmission to the as well (Otten et al., 2018), (Franco et al., 2020), (Zhang et al., 2021), (Franco et al., 2023). So the questi on arises whether ‘destructive’ positive selection may be also acting in the germ sells. In the germline, seminal study (Stewart et al., 2008) established the paradigm of purifying germline selection based on high synonymity of inherited heteroplasmies in the mtDNA ‘mutator' mice, as confirmed recently (Kremer et al., 2025). As for natural mtDNA mutations (as opposed to the mutator mouse mutations), the influential study (Floros et al., 2018) initially asserted purifying germline selection in human primordial germ cells (PGCs) based on an apparent increase in the synonymity of mtDNA mutations in pooled PGCs from early to late stages of development. However, we later found that this (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint synonymity increase was caused by co -amplification of a nuclear pseudogene of mtDNA (NUMT). Opportunely, the authors also published single -PGC mutational data, which were not contaminated by NUMT (Floros et al., 2022). Our analysis of single-PGC data showed preferential expansion of non-synonymous mutations, i.e., demonstrated positive germline selection in PGCs (Fleischmann et al., 2021) , (Fleischmann et al., 2024). Very recently, the discussion of germline selection was supplemented by single -oocyte mutational analysis using high-fidelity, double -stranded deep sequencing, ‘ds - sequencing’ (Arbeithuber et al., 2025). The authors reported, in apparent agreement with Fl oros’, purifying germline selection at higher mutant fractions. This

was of great interest to us as it challenged our criticisms of the Floros et al. study. Furthermore, pilot analysis showed low synonymity of oocyte mutations comparable to that expected from their spectrum without assumptions of selection (0.24). Moreover, synonymity of high-fraction mutations (>0.01) appeared to be even lower (0.21) which challenged purifying selection at higher mutant fractions and hinted of positive selectio n instead. The synonymity results were not statistically significant, however. We thus reanalyzed the Arbeithuber et al. data using a cumulative curve and ‘expansion analysis’ approaches. We found that an impression of purifying selection of coding mutations resulted from comparison to non-coding mutations, which are highly positively selected. After correction for that, the prevailing selection pattern among coding mutations turned out to be positive as well.

True purifying selection of coding mutations or a baseline shift effect caused by comparison to positively selected non-coding mutations? To estimate selection in oocyte mtDNA, Arbeithuber et al. compared the coding -to-noncoding mutation ratio above mutant fraction (MF) of 0.01 to that below 0.01. The expectation was that purifying selection that removes predominantly coding mutations will manifest itself by reducing this ratio. The advantage of this approach is that it does not use the absolute coding vs. non -coding ratio (which depends on non -selection factors, e.g., mutational spectra), but only on the change of the ratio with mutant fraction. Indeed, the ratio was significantly lower for mutations above 0.01, which seemed to naturally “ suggest [purifying] selection operating against coding mutations in oocytes, especially for variants with high frequency” (Arbeithuber et al., 2025). We have previously used a similar approach, i.e., we followed the change of the ratio of non -synonymous to synonymous mutations vs. mutant fraction and in this way demonstrated positive mtDNA selection in single PGCs (Fleischmann et al., 2024). Our approach, similar to the coding vs. non-coding ratio used by Arbeithuber et al. is, in essence, a comparison of selection in ‘tested’ sets of mutations (nonsynonymous or coding, correspondin gly) vs. ‘reference’ sets (synonymous or noncoding). A reference used to measure selection is logically supposed to be free of selection, otherwise selection estimate will certainly be biased. Thus, our concern is that Arbeithuber et al. used as their reference non -coding mutations, which we and others have previously shown to be under strong positive selection in single somatic cells (Nekhaeva et al., 2002), (Korotkevich et al., 2025). If so, then an increased proportion of non -coding mutations at mutant fractions above 0.01 reported by Arbeithuber et al. can be explained by strong positive selection of non -coding mutations in comparison to which coding mutations erroneously appear to be under negative selection, rather than by purifying selection of coding mutations themselves. A way to distinguish between these two alternatives is to compare selection in both coding and non-coding mutations to a neutral reference - i.e., to an established set of non - selected mutations - such as the set of synonymous mutations. Intracellular dynamics of mutations and graphic tool for comparison of selection between mutation types: the cumulative curves. To compare selection between mutation types based on single cell data we considered that the dynamics of mtDNA molecules (including molecules with neutral mutations) in single cell lineage (such as PGC lineage leading to a given oocyte) is primarily driven by intracellular random genetic drift. After each cell division daughter molecules either remain in the lineage that we trace to the oocyte or go into the sister cell and are effectively lost for the traced lineage. Nascent (just generated) neutral muta tions are initially present at low fraction. As PGCs continue to proliferate, most of mutations are lost in driven from the lineage by genetic drift, a low proportion of random ‘lucky’ ones expand into high-fraction clones. For neutral mutations, the growth of large clones is exactly compensated by the random loss of small mutations, which is equivalent to such clonal expansion being non-selective (Coller et al., 2001). In this model, the generation of new mutations during replication of mtDNA creates a continuous flow of mutations from individual nascent to large clones, from low to high mutant fraction. Intracellular mutational dynamics can be graphically presented by cumulative curves Fig.1 ( see Methods and (Franco et al., 2025) for details). Points on the cumulative (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint curve are mutant clones in different cells; each one is essentially a ‘snapshot’ of the clonal expansion. Because mutations were generated continuously, these ‘snapshots’ were taken at different points of expansions, so the entire curve is a collective trajectory of clonal expansions. To the left in the graph are mostly ‘young’ small clones or even nascent individual mutant molecules, on the right – old ‘successful’ mutations, which were lucky to have grown into large clones. Such trajectories can be const ructed separately for different types of mutations. The green curve in Fig. 1 represents the trajectory clonal expansion of synonymous, non-inherited, single oocyte mutations from the dataset (Arbeithuber et al., 2025). Selection by definition manifests itself as systematic acceleration (positive selection) or deceleration of clonal expansion, including complete removal of clones (negative/purifying selection). Simulations (Franco et al., 2025) imply that positive and negative selection cause cumulative curves to deviate upwards and downwards from the neutral drift curve. The real -life dynamics are more complex that the simple genetic drift model presented above. In addition to selection there are changes in the effective population size, most notable once PGC becomes a growing oocyte. Subpopulations on mtDNA (Cote-L’Heureux et al., 2023), (Cote-L’Heureu et al., 2023) are also affecting the dynamics as does potential heterogeneity of cells. These and potentially other deviations from basic drift model may change the shape of the curve and some of them have been already simulated in silico (Franco et al., 2025). In addition the shapes of the curves are likely affected by random variance, especially where datapoint density becomes low. Despite potential complexity of the intracellular dynamics, most factors that determine the curve shapes, are the same for the types of mutations that are compared (e.g. coding vs synonymous), because mutations are sampled from the same cellular pool. Thus when one compares cumulative curve of mutations sampled from the same dataset, the only factor that can drive a systematic difference between the curves is selection. Thus the differences between the curves seen in Fig. 1 demonstrate strong positive selection of non- coding mutations (blue is consistently above green). Coding mutations curve (orange) appear to consistently exceed the synonymous green curve above mutant fraction of about 0.005. Interestingly, the curve for conserved coding mutations appear to exceed the coding mutatiojns curve, as if being conserved (and therefore more detrimental) augments selection. This implies that positive selection of coding mutations may be ‘destructive’, i.e., specifically targeted towards de trimental mutations. Interestingly, when coding conserved mutations are limited to the cytochrome C oxidase subunit 1 gene, the corresponding cumulative curve is below the neutral synonymous curve, indicating a prevalent negative selection in this type of mutations. Whereas cumulative curves in Figure 1 present an intuitive representation of the mtDNA dynamics and selection in the germline, we have not yet developed an approach to determine the statistical significance of these trends. We therefore used a different, m ore quantitative, approach to statistically confirm preliminary observations drawn from Figure1. Relative clonal expansion, 𝓢ₜ, as а metric of intracellular selection. As discussed above, mtDNA mutations are essentially moved to higher fractions by clonal expansion, which is a combination of random drift and selection. Thus a natural way to detect/evaluate selection is to compare the expansion, i.e., the proportion of t otal ‘mutant mass’ (aggregate mutant fraction) that has been ‘moved’ above a given mutant fraction threshold between the tested type of mutations (e.g., coding) and the non-selected, synonymous mutations. Below we formalize this intuitive approach. For the purpose of our analysis, a measure of clonal expansion in a given subset of mutations (e.g. coding) over the threshold t, denoted 𝓔ₜ(subset), is defined as the ‘mutant mass’ (the aggregate mutant fraction) of only those mutant clones whose mutant fractions exceeded a threshold t, normalized by the total aggregate MF of this subset. The value of 𝓔ₜ(subset) itself is not characteristic of the subset of mutations because it depends on many parameters (cell type, age, etc.) In contrast, the ratio 𝓔ₜ(test)/ 𝓔ₜ(synonymous), depends only on selection in the tested subset, because all other parameters are automatically kept equal because mutations of the tested subset of the synonymous mutations are sampled from the same pool of cells. Thus the measure of selection i.e., the acceleration of clone expansion compared to random synonymous Figure 1. Cumulative proportion curves of relevant types of mutations (color coded). (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint expansion is 𝓢ₜ, i.e., the excess of expansion in tested mutations compared to neutral mutations, 𝓔ₜ(tested) - 𝓔ₜ(synonymous), divided by the expansion of neutral mutations, 𝓔ₜ(synonymous). 𝑺ₜ(𝒕𝒆𝒔𝒕𝒆𝒅) = (𝑬𝒕(𝒕𝒆𝒔𝒕𝒆𝒅) − 𝑬𝒕(𝒔𝒚𝒏𝒐𝒏𝒚𝒎)) 𝑬𝒕(𝒔𝒚𝒏𝒐𝒏𝒚𝒎) . 𝓢ₜ>1 implies positive selection, 𝓢ₜ<1 – negative/purifying selection. Statistical significance of selection being present, i.e. 𝓢ₜ being not equal to 1, can be determined using Monte Carlo-style resampling from the set of synonymous mutations (see Methods for details and further justification of the 𝓢ₜ metrics). Non-coding mtDNA mutations are under strong positive selection in oocytes. We first tested our expectation that non-coding mutations in (Arbeithuber et al., 2025) are under positive selection. We started by calculating 𝓢ₜ with t=0.01, following Arbeithuber et al.’s suggestion that 0.01 approximately separates the two different selection regimes. The results (Table 1 , top row) demonstrate highly significant (P ~ 0.0002) and strong ( 𝓢0.01 >3) single-cell positive selection of non -coding mutations, in agreement with past observations (Nekhaeva et al., 2002), (Korotkevich et al., 2025). To show robustness of the result, we confirmed that thresholds ranging from 0.01 to 0.0003 (a generic half - order of magnitude descending geometric series) all support highly significant positive selection (Supplemental Table S1, top row). The differences in selection intensity at different thresholds indicate that selection does vary with mutant fraction. We are exploring this in more detail in an upcoming study (Franco et al., in preparation). Coding mutations are NOT under purifying selection. Instead, selection is positive. Strong positive selection of non -coding mutations implies that the apparent purifying selection of coding mutations reported by Arbeithuber et al. is likely a misreading arising from comparison to a biased reference. To test this possibility, we directly c ompared expansion of coding mutations to that of unselected synonymous mutations (Table 1, second row). Strikingly, this revealed not only the anticipated absence of purifying selection in coding mutations, but also the unexpected highly significant (p ~ 0.002) and moderately strong ( 𝓢0.01~2-fold) positive selection. Thus, the very strong positive selection of non - coding mutations, when they are used as reference, not only ‘makes up’ purifying selection in coding mutations, but also conceals their inherent positive selection. Positive selection of coding and non -synonymous mutations may appear to contradict the finding of Arbeithuber et al. that the MLC concervation score of all oocyte mutations appears to decrease in mutations with MF above 0.01 compared to those below 0.01. This observation was interpreted as evidence for purifying selection. To explain the contradiction, we hypothesized that this effect could be primarily driven by the aggressive expansion of non-coding mutations (which are part of all mutations). Non-coding mutations are expected to be overall less conserved than coding mutations co their selective expansion would cause average conservation score of all mutations to decrease. To test our hypothesis, we used GERP++ conservation metric (Davydov et al., 2010) as a substitute for metrics used by Arbeithuber et al, as we have good understanding of and much experience in working with GERP (see Methods). GERP++ is expected to be generally co -linear with the MLC conservation score used by Arbeithuber et al. Indeed, we, like Arbeithuber at al., also observed a general decrease in average GERP score above MF 0.01 (from 1.9 below to -3.9 above). According t o GERP estimates, non - coding mutations as a group are indeed less conserved (average GERP -6.0), than coding mutations (average GERP +5.6). Thus, the explanation for decreased conservation of mtDNA mutations at higher fractions is as follows: aggressive positive selection of non -coding mutations

in higher prevalence of relatively non -conserved non-coding mutations at higher mutant frac tions which naturally causes an overall reduction in conservation. Indeed, the trend disappears and is actually reversed once non-coding mutations are removed: the a verage GERP score of coding -only mutations increases from 5.5 below MF 0.01 to 13.3 above M F 0.01, in agreement with our finding of positive selection in coding mutations. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint Positive selection in oocytes depends on detrimental effects of mtDNA mutations. The use of a conservation measure helps to address an important conceptual question: what is drives positive selection of coding mutations? Specifically, we asked whether positive selection of coding mutations is related, at least in part, to the detrimen tal effects of these mutations. Of note, a substantial proportion of coding mutations are not detrimental but are essentially neutral. Thus, the subset of conserved coding mutations is expected to be enriched for more detrimental variants. If selection in deed depends on detrimental effects of mutations, then conserved coding mutations should show increased selection, on average. Conversely, if average selection among conserved mutations is not higher than in all mutations, then the hypothesis is likely false. To discern these possibilities, we selected top ~2/3 most conserved coding mutations (i.e., GERP++>10) and repeated the 𝓢100 analysis. As seen ( Table 1, third row), both the intensity and statistical significance of selection increased in conservative coding mutations compared to all coding mutations. This supports the (perhaps counterintuitive) possibility that positive selection of coding mutations in oocytes is driven by the detrimental effects of these mutations. Some mutations are under negative selection. The selection measure 𝓢ₜ used in this study is suited to identifying prevailing selection patterns in sets of multiple mutations (e.g., all non -synonymous or all coding mutations), but is yields no information about individual mutations. Thus some mutations in a set that shows overall positive selection may not be not under selection or even be under negative (purifying) selection. Their effect on the overall selection pattern may be masked by a stronger effect of prevailing positively selected mutations. To demonstrate purifying selection in a dataset such as that of Arbeithuber et al., one needs to identify a subset of mutations in which purifying selection prevails over positive selection; 𝓢ₜ can then detect this negative selection signal. A recent study in mtDNA mutator mice (Kremer et al., 2025) determined that the gene for cytochrome c oxidase subunit 1 (Co1) showed the strongest decrease of non-synonymous mutations in the third generatio n after creation of mutations, suggesting that purifying selection is the strongest in Co1 compared to other mtDNA regions. Thus, we focused our search for negative selection on the Co1 gene. The results of the selection analysis of non -synonymous Co1 mutations are shown in Supplemental Table S1 (bottom row). As expected, they are under prevalent negative (purifying) selection. The significance of this trend is marginal (omnibus p -value ~0.06), which is not surprising given the very small number of data points in the Co1 subset. This implies that not -yet-inherited germline mutations that are under purifying selection indeed exist, but their dynamics is usually not prevailing except in special sunsets of mutations. What about non-synonymous mutations? Usually research on selection operates with synonymity changes as indicators of selection. Following Arbeithuber et al. we concentrated on the comparison of coding, noncoding and conservative mutations rather than traditional nonsynonymous mutations. In fact, nonsynonymous mutations while demonstrating the same trends as coding mutations did not reach statistical significance. This may be related to the lower numbers of non-synonymous mutations. Additionally, our preliminary analysis shows that selection signal a mong nonsynonymous mutations is partially obscured by potential complex subpopulation structure of the intracellular population of the oocyte mtDNA. We are discussing this aspect in our manuscript in preparation. We are confident that there is nothing spec ial about nonsynonymous mutations, and they are also under positive selection. Discussion.

matters: comparing to neutral mtDNA mutations reveals positive selection in the germline. This study was prompted by the apparent contradiction between low synonymity of the oocyte mtDNA mutations and the purifying selection reported in a recent study (Arbeithuber et al., 2025). Purifying selection should have increased the synonymity of mutations, so the observed low levels of synonymity were puzzling. Because in Arbeithuber et al., purifying selection was primarily assessed based on the relative higher prevalence, in oocytes, of non -coding mtDNA mutations over coding mutations, we asked wheth er positive selection of non - coding mutations, rather than negative selection of coding mutations could have caused this effect. This interpretation was based on previous demonstrations that non -coding mutations are indeed prone to positive selection in si ngle cells, at least in somatic tissues (Nekhaeva et al., 2002), (Korotkevich et al., 2024). With this hypothesis in mind, we tested it and, indeed, found strong positive selection of non -coding mtDNA mutations in oocytes ( Table 1 and Table S1 , top row). Positive selection in noncoding mutations may be driven, (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint among other potential mechanisms, by the replication/transcription switch based on structural polymorphism of the G-quadruplex (Wanrooij et al., 2010), (Tan et al., 2016), (Agaronyan et al., 2015), (Gupta et al., 2023), or the second light strand promoter (Tan et al., 2022). These two mechanisms appear to correspond to two blocks of mutations that cause intracellular mtDNA expansion identified in (Nekhaeva et al., 2002): the ‘C-tract block’ and the ‘16028-16050 block’. If non-coding mutations are under positive selection, they cannot be used as an unbiased reference for assessing selection on coding (or any other) mutations; an ideal

should be under no selection. We therefore reanalyzed potential selection in coding mutations using the presumably selectively neutral set of synonymous mutations. As expected, using the synonymous reference, we showed that coding mutations are not under purifying selection. Moreover, we discovered that they are under positive selection: as a set, they expand faster than synonymous mutations, which are driven solely by random drift. To make this even more surprising, the effect of positive selection is augmented among most conservative (i.e., most likely detrimental) non -synonymous mutations. While conceptually counterintuitive, this observation is in line with several prior observati ons of positive selection of detrimental mutations, both in somatic cells and in the germline. The idea of positive selection of detrimental mutations in somatic cells has been widely discussed in the literature since inspirational paper by Aubrey de Grey (De Grey, 1997), who suggested the popular SOS (“survival of the slowest”) model for positive selection. Multiple cases and mechanisms of positive selection of detrimental mutations have been proposed/discussed since then (e.g., SSD – Stochastic Survival of the Densest (Insalata et al., 2022)). For most recent demonstrations of positive selection of detrimental mutations see: (Fleischmann et al., 2024), (Kuiper et al., 2025), (Franco et al., 2025), (Korotkevich et al., 2025). Potential role of the ‘destructive’ positive selection of detrimental mutations. Positive selection of potentially detrimental mutations appears counterproductive and even suicidal, because germ cells with expanded detrimental mtDNA mutations (or their descendants – mutant embryos or even grown up individuals) must eventually be lost, as almost all non - synonymous coding mutations fail to be fixed in the population. “Suicide”, however, may be beneficial if it removes carriers of detrimental mutations. These include non-viable fetuses and severely af fected offspring, which, if not removed, may pose a risk/burden to the mother. Worse still, if such individuals survive to reproductive age, they may disseminate detrimental mutations in the population. In this scenario, positive selection may help push the heteroplasmy of a detrimental mutation to lethal or highly unfit levels, such that carriers of these mutations are removed sooner rather than later to minimize the damage. This scenario would be counterproductive if positive selection operated from very low mutation levels, as it would effectively increase the mutation rate. In contrast, if positive selection were initiated at relatively high mutant fractions, it would targ et specifically the dangerous, “breakthrough” mutation carriers that neutral drift has failed to eliminate. Interestingly, positive germline selection indeed appears to be initiated at higher mutant fractions: our study of mother –child inheritance of detrimental mutations reveals an arch -shaped selection profile, in which selection is absent or even negative at low mutant fractions, becomes positive at intermediate fractions, and then reverts to negative at very high (likely unsustainable) levels (Franco et al., 2023).

Data: data were retrieved following the Supplemental material instructions in (Arbeithuber et al., 2024). Weighted mutant fraction The weighted mutant fraction used here is defined as the fraction of mutant mtDNA molecules among all mtDNA molecules. Note that each “mutation” (a row in the spreadsheet of the Arbeithuber et al. dataset), in fact, represents an intracellular clone of mutant molecules. In the weighted mutant fraction approach, the contribution of each mutation to the aggregate mutation load is proportional to its mutant fraction; it is counted as a clone size, not as a single “mutational event”. As a side note, the conventional use of the number of mutations as a measure of the number of mutational events is problematic, because the number of surviving clones is a complex function of the intensity and duration of random drift and thus depends on s election in a convoluted way. Thus, the “number of mutations” is not a good measure of mutation burden or selection. Weighted synonymity. The measure of synonymity used in this study is the ratio of the number of mtDNA molecules with synonymous mutations to the number of molecules with any coding mutations. As with the weighted mutant fraction, this ratio (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint is based on aggregate mutant fractions rather than on the number of mutations (clones). Note that conventional measures such as hN/hS and related metrics are based on the number of mutations (clones), not on mutant fractions. Historically, numerical synonymity measures were developed for inherited nuclear mutations, which are not measured by fraction, but by homo/heterozygosity. The advantage of weighted synonymity is its ability to assess selection signified by changes in mutant fraction. Conventional numerical synonymity captures only the appearance or disappearance of entire mutant clones, but not changes in clone size. Thus, n umerical synonymity misses the endpoints of selective clonal expansion, which is a very important component of germline selection. Not-yet-inherited (current generation) vs. inherited (previous generation)germline mutations. The mutations discussed in this study are mostly “recent” mutations, i.e., those generated in the current generation. We define these as mutations observed in only one oocyte of a given donor. Being non -repeated across oocytes from the same donor implies that the mutation was unlikely to be present at a significant fraction in the oocyte that eventually developed into the oocyte donor who participated in the study. Otherwise, it would likely appear in more than one oocyte from that donor. Thus, these non-repeated mutations were likely created after the donor’s precursor oocyte became an embryo, implying that they were not inherited from the previous generation. Cumulative mutant proportion curves. The cumulative fraction curves used in the graphical

were adapted (Franco et al., 2025), from the approach developed by Yuan et al. (Yuan et al., 2020). The advantage of these curves as a graphical tool is that they can visually represent neutral drift and are highly sensitive to both positive and negative selection, as demonstrated in simulations (Franco et al., 2025). The precise shape of the cumulative curve of experimental data likely depends, in addition to selection, on many other factors, i ncluding the size of the intracellular population of mtDNA molecules, the number of replications, and more. However complex this parameter landscape is, it is identical for synonymous and nonsynonymous mutations (or coding and noncoding mutations, or else), except for selection. Thus, comparing cumulative curves has the advantage of factoring out multiple difficult -to-account-for parameters and focusing on what matters most. To construct these curves, the mutant fractions are ranged inversely (largest to smallest). Then, the cumulative numerical proportion of all mutations with mutant fraction equal or larger than that of the given mutation to the total mutant load is calcul ated for each mutation. The curve is generated by plotting the mutant fraction of each mutation as a function of the cumulative proportion of the number of mutations with a mutant fraction smaller than the given mutation (in reverse). Then, for better visibility of trends at the high fraction end, where they matter most, each axis is log transformed. Cumulative curves are shown in Figure A of the graphical summary and in Fig .1 of the main text. “Relative expansion” as a measure of selection. What truly implies selection is not the low absolute synonymity per se , but the decrease of synonymity from 0.25 to 0.15 as mutations progress from a lower fraction (e.g., MF 0.01). Accordingly, our assessment of selection is based on comparing the dynamics of different mutation sets (i.e., the ratios of high -fraction to low -fraction mutation load in synonymous vs. non -synonymous mutations), rather than on endpoint composition (e.g., synonymity alone). This “dynamic” approach, which compares low-fraction to high-fraction mutations under the assumption that clonal expansion transforms low -fraction mutations into high - fraction ones, automatically avoids many issues related to spectrum and hotspot correctio n. Low - and high -fraction mutation sets presumably flow one into the other, so all mutations are created under the same initial biases. Thus, when we calculate high -to-low ratios, these biases cancel. We previously used a dynamic approach to assess selection in human PGCs (Fleischmann et al., 2024) and malignant cells (McCastlain et al., 2025). The Relative Expansion analysis used here is also of this “dynamic” type. Of note, the comparison of the coding/noncoding ratio in high -MF vs. low-MF mutations used by Arbeithuber et al. is also a ‘dynamic’ approach. It utilizes, however, a biased reference set and the numerical counts, which disregards expansion of clones beyond crossing a threshold. Although 𝓔ₜ is defined as a ratio of mutational mass above and below a fixed threshold, its interpretation depends on the underlying biological context. In a closed system without mutation input, such a ratio would diverge as all clones eventually cross the threshol d. In contrast, in the germline, de novo mtDNA mutations continuously replenish the pool of small clones, while large clones are lost from the mutational pool upon reaching homoplasmy (or lost to selection). Under these conditions, the system approaches a steady state in which the inflow of new mutations balances the outflow of fixed/lost genotypes. At equilibrium, 𝓔ₜ therefore reflects the effective intensity of clonal expansion across the threshold, rather than an unbounded extent of expansion. (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint Monte Carlo significance of the Relative Expansion selection. We use a non-parametric Monte Carlo simulation to test the statistical significance of selection. The null hypothesis is that there is no selection in the tested set of mutations—i.e., that the tested set is not different from a neutral set of synonymous mutations. To test this null, we randomly sample with replacement from the set of synonymous mutations to generate simulated test sets of the same size as the experimental set and compute the relative expansion for each simulated set. This resampling is repeated mult iple times (20,000 in this study), and the fraction of cases where the relative expansion ratio is equal to or larger than that observed in the experimental test set is taken as the p-value. When selection was estimated at multiple thresholds (typically 0.01, 0.003, 0.001, 0.0003), family-wise adjusted omnibus p-values were computed via Westfall–Young min- p adjustment across thresholds. GERP++ as a conservation score. To test the predictions of the two hypotheses, we used GERP++ (Davydov et al., 2010) as a metric of conservation of mutations. We have previously constructed and validated a particular embodiment of the GERP++ score (K. Gunbin et al., 2016), (K. V. Gunbin et al., 2016), which produces a range of scores between ~-80 for most non-conservative to ~40 for most conservative sites in mtDNA. This continuous metric shows more detail in the mutation conservation landscape than the binary PhastCons score. GERP is s imilar to the mitochondrial local constraint (MLC), except that GERP is based on deeper, interspecies phylogenies, whereas MLC uses more shallow, intraspecies phylogenies. Therefore, GERP is expected to be better at evaluating more conservative (and thus m ore detrimental) mutations. In fact, the GERP embodiment used here is based on particularly deep phylogenies. This is important because one expects first -generation germline mutations that are studied here to be particularly detrimental (compared, e.g., to mutations that achieved homoplasmy, on which MLC is based) (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint

Agaronyan, K., Morozov, Y.I., Anikin, M., Temiakov, D., 2015. Replication-transcription switch in human mitochondria. Science 347, 548–551. https://doi.org/10.1126/science.aaa0986 Arbeithuber, B., Anthony, K., Higgins, B., Oppelt, P., Shebl, O., Tiemann-Boege, I., Chiaromonte, F., Ebner, T., Makova, K.D., 2025. Allele frequency selection and no age-related increase in human oocyte mitochondrial mutations. Sci. Adv. 11, eadw4954. https://doi.org/10.1126/sciadv.adw4954 Arbeithuber, B., Anthony, K., Higgins, B., Oppelt, P., Shebl, O., Tiemann-Boege, I., Chiaromonte, F., Ebner, T., Makova, K.D., 2024. Mitochondrial DNA mutations in human oocytes undergo frequency-dependent selection but do not increase with age. https://doi.org/10.1101/2024.12.09.627454 Bua, E., Johnson, J., Herbst, A., Delong, B., McKenzie, D., Salamat, S., Aiken, J.M., 2006. Mitochondrial DNA-deletion mutations accumulate intracellularly to detrimental levels in aged human skeletal muscle fibers. Am. J. Hum. Genet. 79, 469–480. Coller, H.A., Khrapko, K., Bodyak, N.D., Nekhaeva, E., Herrero-Jimenez, P., Thilly, W.G., 2001. High frequency of homoplasmic mitochondrial DNA mutations in human tumors can be explained without selection. Nature Genetics 28, 147–150. Cote-L’Heureu, A., Franco, M., Maithania, Y.N.K., Popadin, K., Woods, D., Tilly, J., Khrapko, K., 2023. The ‘Stem’ and the ‘Workers’ of the mtDNA population of the cell. Evidence from mutational analysis (preprint). Genomics. https://doi.org/10.1101/2023.04.14.536897 Cote-L’Heureux, A., Maithania, Y.N., Franco, M., Khrapko, K., 2023. Are some mutations more equal than others? eLife 12, e87194. https://doi.org/10.7554/eLife.87194 Davydov, E.V., Goode, D.L., Sirota, M., Cooper, G.M., Sidow, A., Batzoglou, S., 2010. Identifying a high fraction of the human genome to be under selective constraint using GERP++. PLoS Comput Biol 6, e1001025. https://doi.org/10.1371/journal.pcbi.1001025 De Grey, A.D.N.J., 1997. A proposed refinement of the mitochondrial free radical theory of aging. BioEssays 19, 161–166. https://doi.org/10.1002/bies.950190211 Fleischmann, Z., Annis, S., Franco, M., Oreshkov, S., Popadin, K., Woods, D.C., Tilly, J.L., Khrapko, K., 2021. Nuclear DNA- encoded fragments of mitochondrial DNA (mtDNA) confound analysis of selection of mtDNA mutations in human primordial germ cells (preprint). Genomics. https://doi.org/10.1101/2021.10.18.464832 Fleischmann, Z., Cote-L’Heureux, A., Franco, M., Oreshkov, S., Annis, S., Khrapko, M., Aidlen, D., Popadin, K., Woods, D.C., Tilly, J.L., Khrapko, K., 2024. Reanalysis of mtDNA mutations of human primordial germ cells (PGCs) reveals NUMT contamination and suggests that selection in PGCs may be positive. Mitochondrion 74, 101817. https://doi.org/10.1016/j.mito.2023.10.005 Floros, V.I., Pyle, A., Dietmann, S., Wei, W., Tang, W.C.W., Irie, N., Payne, B., Capalbo, A., Noli, L., Coxhead, J., Hudson, G., Crosier, M., Strahl, H., Khalaf, Y., Saitou, M., Ilic, D., Surani, M.A., Chinnery, P.F., 2018. Segregation of mitochondrial DNA heteroplasmy through a developmental genetic bottleneck in human embryos. Nature Cell Biology 2018 20:2 20, 144–151. Floros, V.I., Pyle, A., Dietmann, S., Wei, W., Tang, W.W.C., Irie, N., Payne, B., Capalbo, A., Noli, L., Coxhead, J., Hudson, G., Crosier, M., Strahl, H., Khalaf, Y., Saitou, M., Ilic, D., Surani, M.A., Chinnery, P.F., 2022. Author Correction: Segregation of mitochondrial DNA heteroplasmy through a developmental genetic bottleneck in human embryos. Nat Cell Biol. https://doi.org/10.1038/s41556-022-01046-z Franco, M., Bandell, J., Popadin, K., Woods, D.C., Khrapko, K., 2025. Are mitochondrial DNA mutations under purifying selection in somatic tissues? https://doi.org/10.1101/2025.10.26.684677 Franco, M., Cote-L’Heureux, A., Fleischmann, Z., Chen, Z., Khrapko, M., Vyshedskiy, B., Braverman, M., Popadin, K., Pickett, S., Woods, D.C., Tilly, J.L., Turnbull, D., Khrapko, K., 2023. A novel, wave-shaped profile of germline selection of pathogenic mtDNA mutations is discovered by bypassing a classical statistical bias. https://doi.org/10.1101/2023.11.21.568140 Franco, M., Fleischmann, Z., Wasco, K., DeRosa, S., Payne, C., Annis, S., Khrapko, K., 2019. Mitochondrial DNA replication-related Nucleotide Patterns: Ancient ruins or living structures? The FASEB Journal 33, lb173–lb173. https://doi.org/10.1096/fasebj.2019.33.1_supplement.lb173 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint Franco, M., Pickett, S., Fleischmann, Z., Khrapko, M., Annis, S., woods, dori, Markuzon, N., Turnbull, D., Khrapko, K., 2020. Can detrimental mtDNA mutations be under positive selection in the germline? The FASEB Journal 34, 1– 1. https://doi.org/10.1096/fasebj.2020.34.s1.09461 Gunbin, K., Peshkin, L., Popadin, K., Annis, S., Ackermann, R.R., Khrapko, K., 2016. Integration of mtDNA pseudogenes into the nuclear genome coincides with speciation of the human genus. A hypothesis. Mitochondrion. http://dx.doi.org/10.1016/j.mito.2016.12.001 Gunbin, K.V., Peshkin, L., Popadin, K., Annis, S., Ackermann, R.R., Khrapko, K., 2016. The time of integration of the human mitochondrial pseudogenes (NUMTs) into the nuclear genome. (Submitted). Data in Brief. Gupta, R., Kanai, M., Durham, T.J., Tsuo, K., McCoy, J.G., Kotrys, A.V., Zhou, W., Chinnery, P.F., Karczewski, K.J., Calvo, S.E., Neale, B.M., Mootha, V.K., 2023. Nuclear genetic control of mtDNA copy number and heteroplasmy in humans. Nature 620, 839–848. https://doi.org/10.1038/s41586-023-06426-5 Insalata, F., Hoitzing, H., Aryaman, J., Jones, N.S., 2022. Stochastic survival of the densest and mitochondrial DNA clonal expansion in aging. Proc. Natl. Acad. Sci. U.S.A. 119, e2122073119. https://doi.org/10.1073/pnas.2122073119 Korotkevich, E., Conrad, D.N., Gartner, Z.J., O’Farrell, P.H., 2025. Selfish mutations promote age-associated erosion of mtDNA integrity in mammals. Nat Commun 16, 5435. https://doi.org/10.1038/s41467-025-60477-y Korotkevich, E., Conrad, D.N., Gartner, Z.J., O’Farrell, P.H., 2024. Selection promotes age-dependent degeneration of the mitochondrial genome. https://doi.org/10.1101/2024.09.27.615276 Kraytsberg, Y., Kudryavtseva, E., McKee, A.C., Geula, C., Kowall, N.W., Khrapko, K., 2006. Mitochondrial DNA deletions are abundant and cause functional impairment in aged human substantia nigra neurons. Nature Genet 38, 518– 520. Kremer, L.S., Golder, Z., Barton-Owen, T., Papadea, P., Koolmeister, C., Chinnery, P.F., Larsson, N.-G., 2025. The bottleneck for maternal transmission of mtDNA is linked to purifying selection by autophagy. Sci. Adv. 11, eaea4660. https://doi.org/10.1126/sciadv.aea4660 Kuiper, L.M., Shi, W., Verlouw, J.A.M., Hong, Y.S., Arp, P., Puiu, D., Broer, L., Xie, J., Newcomb, C., Rich, S.S., Taylor, K.D., Rotter, J.I., Bader, J.S., Guallar, E., Van Meurs, J.B.J., Arking, D.E., 2025. Deleterious mitochondrial heteroplasmies exhibit increased longitudinal change in variant allele fraction. iScience 28, 112590. https://doi.org/10.1016/j.isci.2025.112590 McCastlain, K., Welsh, C., Ni, Y., Ding, L., Chang, T.-C., Autry, R.J., Sejdiu, B.I., Pan, Q., Franco, M., Chen, W., Wu, H., Gonzalez-Pena, V., Schreiner, P., Arunachalam, S., Joo, J.H., Brady, S., Zhang, J., Gawad, C., Evans, W.E., Babu, M.M., Khrapko, K., Yu, J., Wu, G., Pounds, S., Kundu, M., 2025. Somatic mtDNA mutations at intermediate levels of heteroplasmy are a source of functional heterogeneity among primary leukemic cells. Sci. Adv. 11, eadt3873. https://doi.org/10.1126/sciadv.adt3873 Mikhailova, A.G., Mikhailova, A.A., Ushakova, K., Tretiakov, E.O., Iliushchenko, D., Shamansky, V., Lobanova, V., Kozenkov, I., Efimenko, B., Yurchenko, A.A., Kozenkova, E., Zdobnov, E.M., Makeev, V., Yurov, V., Tanaka, M., Gostimskaya, I., Fleischmann, Z., Annis, S., Franco, M., Wasko, K., Denisov, S., Kunz, W.S., Knorre, D., Mazunin, I., Nikolaev, S., Fellay, J., Reymond, A., Khrapko, K., Gunbin, K., Popadin, K., 2022. A mitochondria-specific mutational signature of aging: increased rate of A > G substitutions on the heavy strand. Nucleic Acids Research 50, 10264–10277. https://doi.org/10.1093/nar/gkac779 Nekhaeva, E., Bodyak, N.D., Kraytsberg, Y., McGrath, S.B., van Orsouw, N.J., Pluzhnikov, A., Wei, J.Y., Vijg, J., Khrapko, K., 2002. Clonally expanded mtDNA point mutations are abundant in individual cells of human tissues. Proceedings of the National Academy of Sciences 99, 5521–5526. https://doi.org/10.1073/pnas.07267019 Nicholas, A., Kraytsberg, Y., Guo, X., Khrapko, K., 2009. On the timing and the extent of clonal expansion of mtDNA deletions: evidence from single-molecule PCR. Exp Neurol 218, 316–319. Otten, A.B.C., Sallevelt, S.C.E.H., Carling, P.J., Dreesen, J.C.F.M., Drüsedau, M., Spierts, S., Paulussen, A.D.C., de Die- Smulders, C.E.M., Herbert, M., Chinnery, P.F., Samuels, D.C., Lindsey, P., Smeets, H.J.M., 2018. Mutation-specific effects in germline transmission of pathogenic mtDNA variants. Hum Reprod 33, 1331–1341. https://doi.org/10.1093/humrep/dey114 Popadin, K., Gunbin, K., Peshkin, L., Annis, S., Fleischmann, Z., Franco, M., Kraytsberg, Y., Markuzon, N., Ackermann, R.R., Khrapko, K., 2022. Mitochondrial Pseudogenes Suggest Repeated Inter-Species Hybridization among Direct Human Ancestors. Genes 13, 810. https://doi.org/10.3390/genes13050810 Serrano, I.M., Hirose, M., Valentine, C.C., Roesner, S., Schmidt, E., Pratt, G., Williams, L., Salk, J., Ibrahim, S., Sudmant, P.H., 2024. Mitochondrial haplotype and mito-nuclear matching drive somatic mutation and selection throughout ageing. Nat Ecol Evol 8, 1021–1034. https://doi.org/10.1038/s41559-024-02338-3 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint Stewart, J.B., Freyer, C., Elson, J.L., Wredenberg, A., Cansu, Z., Trifunovic, A., Larsson, N.G., 2008. Strong purifying selection in transmission of mammalian mitochondrial DNA. PLoS Biol 6, e10. Tan, B.G., Mutti, C.D., Shi, Y., Xie, X., Zhu, X., Silva-Pinheiro, P., Menger, K.E., Díaz-Maldonado, H., Wei, W., Nicholls, T.J., Chinnery, P.F., Minczuk, M., Falkenberg, M., Gustafsson, C.M., 2022. The human mitochondrial genome contains a second light strand promoter. Molecular Cell 82, 3646-3660.e9. https://doi.org/10.1016/j.molcel.2022.08.011 Tan, B.G., Wellesley, F.C., Savery, N.J., Szczelkun, M.D., 2016. Length heterogeneity at conserved sequence block 2 in human mitochondrial DNA acts as a rheostat for RNA polymerase POLRMT activity. Nucleic Acids Res 44, 7817– 7829. https://doi.org/10.1093/nar/gkw648 Wanrooij, P.H., Uhler, J.P., Simonsson, T., Falkenberg, M., Gustafsson, C.M., 2010. G-quadruplex structures in RNA stimulate mitochondrial transcription termination and primer formation. Proc. Natl. Acad. Sci. U.S.A. 107, 16072–16077. https://doi.org/10.1073/pnas.1006026107 Wei, W., Tuna, S., Keogh, M.J., Smith, K.R., Aitman, T.J., Beales, P.L., Bennett, D.L., Gale, D.P., Bitner-Glindzicz, M.A.K., Black, G.C., Brennan, P., Elliott, P., Flinter, F.A., Floto, R.A., Houlden, H., Irving, M., Koziell, A., Maher, E.R., Markus, H.S., Morrell, N.W., Newman, W.G., Roberts, I., Sayer, J.A., Smith, K.G.C., Taylor, J.C., Watkins, H., Webster, A.R., Wilkie, A.O.M., Williamson, C., Ashford, S., Penkett, C.J., Stirrups, K.E., Rendon, A., Ouwehand, W.H., Bradley, J.R., Raymond, F.L., Caulfield, M., Turro, E., Chinnery, P.F., 2019. Germline selection shapes human mitochondrial DNA diversity. Science 364, eaau6520. Yuan, Y., Ju, Y.S., Kim, Y., Li, J., Wang, Y., Yoon, C.J., Yang, Y., Martincorena, I., Creighton, C.J., Weinstein, J.N., Xu, Y., Han, L., Kim, H.-L., Nakagawa, H., Park, K., Campbell, P.J., Liang, H., Consortium, P., 2020. Comprehensive molecular characterization of mitochondrial genomes in human cancers. Nature Genetics 52, 342–352. Zhang, H., Esposito, M., Pezet, M.G., Aryaman, J., Wei, W., Klimm, F., Calabrese, C., Burr, S.P., Macabelli, C.H., Viscomi, C., Saitou, M., Chiaratti, M.R., Stewart, J.B., Jones, N., Chinnery, P.F., 2021. Mitochondrial DNA heteroplasmy is modulated during oocyte development propagating mutation transmission. Sci. Adv. 7, eabi5657. https://doi.org/10.1126/sciadv.abi5657 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted January 9, 2026. ; https://doi.org/10.64898/2025.12.31.697248doi: bioRxiv preprint